Image labeler Train cascade error!

Hi,

I have a set of labeled images using the image labeler app. I have exported using the builtin function to gndTruth file.

I have the following code:

load('gndTruth.mat');
positiveInstances = gndTruth(:,2:1);
imDir = fullfile('positveImgDir');
addpath(imDir);

negativeFolder = fullfile('NegativeImgDir');
negativeImages = imageDatastore(negativeFolder);

trainCascadeObjectDetector('bluesDetector.xml',positiveInstances, ...
    negativeFolder,'FalseAlarmRate',0.1,'NumCascadeStages',5);

I have no idea why it is causing error.

Answer 1

[Ridwan Alam]

Could the error be that I'm using pixel label instead of line or rectangle label?

Yes, sorry!

Update:

[imds,pxds] = pixelLabelTrainingData(gTruth);

But I don't think you can use imds and pxds for trainCascadeObjectDetector().

Because the positiveInstances variable requires the second table column or structure field to contain an M-by-4 matrix of M bounding boxes. Each bounding box is in the format [x y width height] and specifies an object location in the corresponding image.

[Ridwan Alam]

I believe you need gTruth created with box labels. Then you can use objectDetectorTrainingData() and trainCascadeObjectDetector().